New features for population design evaluation and optimisation with R functions: PFIM Interface 3.1 and PFIM 3.2
Caroline Bazzoli (1), Thu-Thuy Nguyen (2), Emanuelle Comets (2), Anne Dubois (2), Hervé Le Nagard (2), France Mentré (2)
(1) Laboratoire Jean Kuntzmann, Département Statistique, Grenoble, France ; (2) INSERM U738 and Université Paris Diderot, Paris, France;
Objectives: To develop the free R function PFIM [1] for population design evaluation and optimization and to illustrate the use of PFIM Interface 3.1 and PFIM 3.2.
Methods: Compared to PFIM Interface 2.1, the new PFIM Interface version 3.1, dedicated to design evaluation and optimisation for multiple response models, incorporates the features that were previously released in version 3.0 of PFIM [2]. Furthermore, the library of "classical" pharmacokinetic (PK) models has been completed by the three compartment models and a library of pharmacodynamic (PD) models is now available. PFIM Interface 3.1 can handle either a block diagonal Fisher matrix or the complete one. Regarding the R scripts version, the key new feature of PFIM 3.2 is the computation of the Fisher information matrix for models including fixed effects for the influence of discrete covariates on the parameters [3] and/or inter-occasion variability (IOV) [4]. The predicted power of the Wald test for a given distribution of a discrete covariate as well as the number of subjects needed to achieve a given power can be computed. Examples of the use of both versions of PFIM are presented in the context of warfarin PKPD.
Results: We used the standard example of warfarin PKPD where warfarin is administered as a single oral dose to 32 subjects. Plasma concentration and effect on prothrombin complex activity (PCA) are measured. A one compartment PK model with first order absorption and elimination is used and the effect on PCA is described by a turnover model with inhibition of the input. First, using PFIM Interface 3.1, we evaluated the empirical rich design and compared it to a design optimised using the Fedorov-Wynn algorithm. With 2.1 less samples than the empirical design, the optimal design provides similar predicted standard errors for the fixed effects. Then, we wanted to evaluate designs with a genetic covariate effect on clearance using PFIM 3.2. With the optimal design and a clearance assumed to decrease by 50% for patients with a mutant genotype, only 8 subjects are needed to obtain a power of 90% for the comparison test detecting the genetic effect. Finally, we planned a crossover PK study to assess the absence of interaction of drug X on warfarin absorption rate-constant (ka).
Conclusions: We illustrated the use of the new versions PFIM 3.2 and PFIM Interface 3.1. They are great tools to evaluate and/or optimise designs for multiple response models and for more complex models quantifying the influence of discrete covariate and/or inter-occasion variability.
References:
[1] http://www.pfim.biostat.fr/
[2] Bazzoli C, Retout S and Mentré F. Design evaluation and optimisation in multiple response nonlinear mixed effect models: PFIM 3.0. Computer Methods and Programs in Biomedicine. 98: 55-65, 2010.
[3] Retout S, Comets E, Samson A and Mentré F. Design in nonlinear mixed effects models: Optimization using the Federov-Wynn algorithm and power of the Wald test for binary covariates. Statistics in Medicine. 26:5162-5179, 2007.
[4] Nguyen TT, Bazzoli C and Mentré F. Design evaluation and optimization in crossover pharmacokinetic studies analysed by nonlinear mixed effects models: application to bioequivalence or interaction trials. American Conference in Pharmacometrics, Mashantucket, United-States, October 4-7, 2009 (poster).
